(543a) A Multi-Fidelity, Physics-Informed Approach to Active Learning-Guided Experiment Design for the Study of Ammonia Synthesis Via Plasma Catalysis

Low temperature plasmas (LTPs) have shown promise for electrification of the chemical industry [1], e.g., for NO_x synthesis, methane reforming and NH₃ synthesis [2,3,4], wherein reactions can happen under room temperatures and pressures via vibrationally excited species that are generated by renewable electricity. LTP-driven chemical synthesis can be enhanced by introducing catalysis [2]. The synergies between plasma and catalysts, such as enhanced electric field, micro-discharges and species impacts on catalyst surface, are postulated to influence productivity and selectivity of chemical synthesis [1]. Nonetheless, the plasma-catalyst interactions are intricate and far from well-understood [1,5]. To this end, a key challenge arises from lack of systematic approaches to explore the synergistic impacts of LTP behavior and catalyst properties on the plasma-catalyst interactions. In this talk, we will present an active learning (AL) framework for the design of experiments with multiple objectives to investigate plasma-catalyst interactions in ammonia synthesis.

AL methods have emerged as a useful tool to address scientific and engineering problems that require intensive and expensive evaluations (experiments or simulations) (see, e.g., [6,7,8]). AL concerns with systematically querying samples from an experimental system (or computational model) to train a data-driven model that correlates design parameters to a performance measure of interest. By repeating the sample query and data-driven model training, AL will allow for efficient exploration of a design space towards optimizing the performance measure. A popular approach to AL is Bayesian optimization (BO), a data-driven optimization method that is particularly suited for optimizing blackbox and noisy objectives [9]. In this talk, we will present a multi-fidelity BO (MFBO) [10] approach to AL based on physics-based models. MFBO leverages different levels of physics-based fidelity for active exploration of the design space, which can significantly accelerate AL when dealing with expensive physics-based models.

In this work, the proposed MFBO framework relies on a 0D model of plasma kinetics combined with density functional theory (DFT) to describe surface reactions on a catalyst. The objective is to identify the favorable plasma parameters that enhance NH₃ production rate and concentration. In particular, we focus on electric field, gas temperature and electron density as the key plasma parameters. Electric field plays an important role in boosting surface reactions by altering the activation energies of these reactions [3,11], while gas temperature and electron density influence the density of reactive species such as H atom, N atom, excited H₂ and N₂ in the plasma chemistry [12]. Here, the DFT calculations, performed in CP2K [13], are used to compute the activation energies of surface reactions under different electric fields, which influence the surface reaction rate coefficients. A kinetic model in ZDPlasKin [14] uses these rate coefficients to describe the plasma-catalytic NH₃ synthesis process. We will demonstrate how the MFBO allows us to systematically and efficiently explore the design space of the aforementioned plasma parameters towards maximizing production rate and concentration of the produced NH₃ despite the fact that exhaustive exploration of the design space can quickly become prohibitive due to the expensive DFT calculations. In MFBO, we leverage two different levels of physics-based fidelity to calculate activation energies, which are Bell–Evans–Polanyi (BEP) principle [15] and transition state theory (TST) [16]. BEP can efficiently predict activation energies based on the reaction energy of the same reaction, constituting a low-fidelity model, whereas TST can accurately calculate the activation energies by an expensive search of transition state species on the catalyst surface. We will show how the MFBO aids in discovering the important surface reactions and the favorable reaction pathways for ammonia synthesis. The proposed physics-informed AL framework can be readily adapted for other plasma catalysis processes.

References

[1] Bogaerts, A., Tu, X., Whitehead, J. C., Centi, G., Lefferts, L., Guaitella, O., ... & Carreon, M. (2020). The 2020 plasma catalysis roadmap. Journal of Physics D: Applied Physics, 53(44), 443001.

[2] Patil, B. S., Cherkasov, N., Lang, J., Ibhadon, A. O., Hessel, V., & Wang, Q. (2016). Low temperature plasma-catalytic NOx synthesis in a packed DBD reactor: Effect of support materials and supported active metal oxides. Applied Catalysis B: Environmental, 194, 123-133.

[3] Che, F., Gray, J. T., Ha, S., & McEwen, J. S. (2017). Improving Ni catalysts using electric fields: a DFT and experimental study of the methane steam reforming reaction. ACS Catalysis, 7(1), 551-562. [4] Bal, K. M., Huygh, S., Bogaerts, A., & Neyts, E. C. (2018). Effect of plasma-induced surface charging on catalytic processes: Application to CO2 activation. Plasma Sources Science and Technology, 27(2), 024001.

[5] Neyts, E. C., Ostrikov, K., Sunkara, M. K., & Bogaerts, A. (2015). Plasma catalysis: synergistic effects at the nanoscale. Chemical Reviews, 115(24), 13408-13446.

[6] Wakabayashi, Y. K., Otsuka, T., Krockenberger, Y., Sawada, H., Taniyasu, Y., & Yamamoto, H. (2019). Machine-learning-assisted thin-film growth: Bayesian optimization in molecular beam epitaxy of SrRuO3 thin films. APL Materials, 7(10), 101114.

[7] Osada, K., Kutsukake, K., Yamamoto, J., Yamashita, S., Kodera, T., Nagai, Y., ... & Ujihara, T. (2020). Adaptive Bayesian optimization for epitaxial growth of Si thin films under various constraints. Materials Today Communications, 25, 101538.

[8] Shao, K., Pei, X., Graves, D. B., Mesbah, A. (2022). "Active learning-guided exploration of parameter space of air plasmas to enhance the energy efficiency of NO_x production." [Manuscript submitted for publication]. Plasma Sources Science and Technology.

[9] Frazier, P. I. (2018). A tutorial on Bayesian optimization. ArXiv Preprint ArXiv:1807.02811.

[10] Song, J., Chen, Y., & Yue, Y. (2019, April). A general framework for multi-fidelity Bayesian optimization with Gaussian processes. In The 22nd International Conference on Artificial Intelligence and Statistics (pp. 3158-3167). PMLR.

[11] Che, F., Gray, J. T., Ha, S., Kruse, N., Scott, S. L., & McEwen, J. S. (2018). Elucidating the roles of electric fields in catalysis: A perspective. ACS Catalysis, 8(6), 5153-5174.

[12] Wang, Y., Craven, M., Yu, X., Ding, J., Bryant, P., Huang, J., & Tu, X. (2019). Plasma-enhanced catalytic synthesis of ammonia over a Ni/Al2O3 catalyst at near-room temperature: insights into the importance of the catalyst surface on the reaction mechanism. ACS Catalysis, 9(12), 10780-10793.

[13] Kühne, T. D., Iannuzzi, M., Del Ben, M., Rybkin, V. V., Seewald, P., Stein, F., ... & Hutter, J. (2020). CP2K: An electronic structure and molecular dynamics software package-Quickstep: Efficient and accurate electronic structure calculations. The Journal of Chemical Physics, 152(19), 194103.

[14] Pancheshnyi, S., Eismann, B., Hagelaar, G., & Pitchford, L. (2008). ZDPlasKin: a new tool for plasmachemical simulations. Bulletin of the American Physical Society, 53.

[15] Vinu, R., & Broadbelt, L. J. (2012). Unraveling reaction pathways and specifying reaction kinetics for complex systems. Annual Review of Chemical and Biomolecular Engineering, 3, 29-54.

[16] Pechukas, P. (1981). Transition state theory. Annual Review of Physical Chemistry, 32(1), 159-177.